poverty and policy selectivity of world bank trust …documents.worldbank.org/curated/en/... ·...

Policy Research Working Paper 7731

Poverty and Policy Selectivity of World Bank Trust Funds

Vera Eichenauer Stephen Knack

Development Research GroupHuman Development and Public Services TeamJune 2016

WPS7731P

ublic

Dis

clos

ure

Aut

horiz

edP

ublic

Dis

clos

ure

Aut

horiz

edP

ublic

Dis

clos

ure

Aut

horiz

edP

ublic

Dis

clos

ure

Aut

horiz

ed

Produced by the Research Support Team

Abstract

The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent.

Policy Research Working Paper 7731

This paper is a product of the Human Development and Public Services Team, Development Research Group. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at [email protected].

Over the past decade, donors of foreign aid quadrupled their annual contributions to trust funds at the World Bank. This earmarking of contributions to donors’ preferred recipient countries and issues has raised concerns about the alignment of trust funds with the performance-based allocations of aid by the International Development Asso-ciation, the World Bank’s concessional lending arm, and raises the question of the role of this new “multi-bi” aid channel. This study finds that the cross-country allocations of aggregate trust fund aid are poverty- and policy-selective. In this respect, they are much more similar to allocations from the International Development Association than from bilateral donors. The allocations of trust fund types

that are more closely controlled by donor countries—recipient-executed and single-donor trust funds—are more strongly related to the strategic interests of donor countries than trust fund aid in general. Trust funds for health and education aid are poverty selective and posi-tively correlated with the World Bank’s assessment of the quality of countries’ sector policies, while environmental trust funds are neither poverty selective nor correlated with the assessed quality of countries’ environmental poli-cies. Overall, the evidence indicates that multi-bi funds administered by the World Bank do not undermine the International Development Association’s allocation criteria.

Poverty and Policy Selectivity of World Bank Trust Funds

Vera Eichenauer (Heidelberg University)

Stephen Knack (World Bank)

JEL: O19

Keywords: foreign aid, World Bank, trust funds, aid allocation, aid selectivity

Acknowledgements: We are grateful for helpful comments on previous and quite distinct versions

by Christopher Kilby, Dan Honig, Andreas Fuchs, Stephan Klingebiel, Axel Dreher, Barbara

Dluhosch, and participants at the 8th Annual Conference on the Political Economy of International

Organizations and at the Trust Fund Workshop at the University of Zurich. We thank Xu Lu for

valuable research assistance, Christopher Kilby for sharing data on important UN votes according

to the US State Department, and Buenaflora P. Cabanela and other staff at the Trust Funds and

Partnerships Department (DFPTF) at the World Bank for providing and explaining the data to us.

Vera Eichenauer gratefully acknowledges financial support from the Swiss Network for

International Studies (SNIS). Much of the content in this paper was included in a broader paper

titled “ ‘Bilateralizing’ multilateral aid? The political economy of World Bank trust funds” and

previously circulated under the title “Trust Fund disbursements to developing countries: who

benefits and why?”

2

1. Introduction

Multilateral trust funds at the World Bank and at other international organizations are increasingly

popular with donors of foreign aid. These new funding modalities allow donor governments to

cooperate with like‐minded donors only, and to earmark their aid for specific countries, sectors or

objectives (as with other bilateral aid), while using the financial and implementation infrastructure

of the multilateral organization hosting them.1 The increasing importance of this “multi‐bi” aid

channel raises questions about its distinctiveness in relation to unearmarked multilateral aid and

other bilateral aid, particularly with respect to country allocation patterns. Trust funds could

potentially undermine the policy‐ and poverty‐selectivity of the World Bank’s IDA (International

Development Association) aid, if donors’ earmarked contributions come partly at the expense of

their core contributions to IDA, and if trust fund allocations are not sensitive to recipients’ quality

of governance or income levels.

In this paper, we exploit a novel and rich data set on World Bank trust fund disbursements

from 2002 to 2012 to analyze the determinants of trust fund allocations across countries,

differentiating trust funds by type and sector. We specifically compare the policy‐ and poverty‐

selectivity of multi‐bi aid to that of other bilateral aid and of the World Bank’s IDA disbursements.

We find that trust funds are similar to IDA, and different from other bilateral aid, in being highly

sensitive to recipient’s income and quality of governance. Multi‐bi funds administered by the World

Bank thus do not appear to undermine IDA’s allocation criteria, even if donors’ contributions to

them partially crowd out their IDA contributions. We leave for future research the question of

whether the increase in trust funds has had an impact on core contributions to IDA,2 their potential

effects on other aspects of World Bank operational policies,3 and the effectiveness of trust funds

relative to traditional bilateral and multilateral core aid.

The remainder of the paper is structured as follows. Section 2 provides background

information on World Bank trust funds and reviews related literature. Section 3 presents some

theoretical arguments concerning donor motives on allocation of trust funds. Section 4 describes the

data, while section 5 discusses the methods and presents the main results for aggregate and sector‐

specific disbursements. Section 6 concludes.

1 Because trust fund contributions are earmarked, they are officially classified by the OECD DAC as bilateral aid,

although multilaterals are the implementing agencies. 2 Eichenauer and Hug (2014) propose a model with a multilateral agent governed by multiple principals with

heterogeneous preferences to better understand the tradeoffs donors face when choosing bilateral, multilateral or trust

fund aid. Assuming a fixed aid budget, their findings suggest that the possibility of trust fund contributions decreases

bilateral and multilateral ‘core’ aid in most cases. During recent IDA replenishment negotiations, the World Bank has

placed a moratorium on solicitations of trust funds from donors by its staff members, indicating that it believes trust

funds can (partially) displace core contributions. 3 Reinsberg (2015) discusses implications for the World Bank.

3

2. Background and Related Literature

Over the last decade, trust funds at the World Bank have proliferated. The total number of active

trust funds considered in this paper exceeded 1,200, excluding the largely independent financial

intermediary funds4 (FIFs) and trust funds at the International Financial Corporation (IFC). These

funds disbursed more than 22.5 billion USD over the fiscal years 2002‐2012, our sample period.5 To

put this number into perspective, gross disbursements by IDA, the fund accounting for most

concessional lending by the World Bank, amounted to 10.9 billion USD in the 2012 fiscal year. Figure

1 shows the significant increase in trust fund disbursements over the last decade and how its volume

compares to IDA funds. Most disbursements were made to specific countries, while smaller shares

are allocated to a region or for global goods. Donors are very heterogeneous in their use of trust

funds to deliver their foreign aid. In the 2002‐2012 sample period, the largest sovereign contributors

to World Bank funds, excluding FIFs, were the United Kingdom, the European Commission, the

United States, and the Netherlands.6

According to the official narrative, trust funds fill ‘gaps in the multilateral system’ with

respect to global public goods and following “emergencies such as natural disasters, disease

outbreaks, and the end of armed conflict, where donors want to coordinate their bilateral aid and

where the MDBs do not grant resources to engage on a sufficient scale” (IEG 2011a: 5). However, a

systematic evaluation of the cross‐country allocation patterns of trust fund aid is still lacking. In this

paper, we focus on the criteria of policy selectivity and poverty selectivity that largely guide IDA

allocations, and that reflect a widespread belief about country characteristics conducive to the

productive use of aid funds.

Three actors are involved in the recent trust fund trend: donor countries, the World Bank,

and recipient countries. Donor governments may use trust funds to target their foreign aid to

priority countries and development issues, while, and in contrast to bilateral aid, delegating

responsibility for its management and implementation to the multilateral organization. Evidence

collected by the Bank’s evaluation unit (IEG 2011a) through structured interviews with 55 officials

of eight donor countries finds that six out of eight donor countries use trust funds to target priority

issues or countries. From the perspective of the multilateral organization, trust funds allow

expanding its global role and operations, and increasing its staff and assets under management.

From a recipient country perspective, trust funds may have several positive implications. For

middle‐income countries seeking technical assistance but are reluctant to borrow for this purpose,

trust funds make technical assistance available at grant terms (IEG 2011a: 7). Trust funds have also

supported post‐conflict and post‐disaster countries and territories that are ineligible to borrow from

the IDA or the International Bank for Reconstruction and Development (IBRD) (e.g., Timor‐Leste,

Aceh in Indonesia, West Bank and Gaza). Moreover, trust funds have encouraged the provision of

4 Programs funded from FIF are typically monitored by separate agencies, such as GAVI, the Vaccine Alliance, or the

Global Fund, and not by World Bank staff. The Bank mainly provides financial administration for FIFs. 5 The fiscal year at the World Bank runs from July 1 to June 30. 6 This information is based on contributions data with donor‐specific information.

4

global public goods (IEG 2011a: viii). While multi‐donor trust funds could, theoretically, improve

donor coordination prior to implementation, reducing excessive and harmful donor fragmentation

in the field (Huq 2010, IEG 2011a: ix), its effects on donor harmonization seem ambiguous (IEG

2011a: 43, Barakat et al. 2012: 34f.) as trust funds usually do not replace existing bilateral and

multilateral projects (Barakat 2009: 112).

An evaluation of World Bank trust funds (IEG 2011a: 7) highlights their importance for

countries in arrears or entities, where IDA is legally forbidden to engage and bilateral donors prefer

not to engage alone. Looking at aggregate official aid flows, however, it is still unclear whether, and

in what sectors or countries, trust fund aid substitutes for (i.e., “crowds out”) or complements

multilateral or bilateral aid, or if it is additional to traditional aid. Using data on donors organized

in the OECD’s Development Assistance Committee (DAC), Reinsberg et al. (2015) find some

evidence that earmarked or multi‐bi aid is additional to “core” multilateral aid, although their

estimations may suffer from potential reverse causality and simultaneity. Based on data for the

World Bank through 2009, Huq (2010) finds that the sectoral allocation of trust funds executed by

recipients is “aligned” (or positively correlated) with IDA but not with IBRD disbursements.

Donor countries might prefer using trust funds rather than bilateral or multilateral aid for a

number of reasons. Donors might seek to supplement IDA funding for particular countries, sectors

and/or projects, and trust funds are often used to co‐finance IDA projects. In other cases, bilateral

donors may use trust funds to complement their bilateral programming when their aid agencies do

not have sufficient presence or expertise in countries to implement programs effectively (OECD

2010: 40, 2011: 29). Donor officials state that the proliferation of trust fund aid is associated with a

need for a new type of aid that is complementary to the existing multilateral and bilateral aid. They

claim that multilateral core contributions to the World Bank and other MDBs cannot achieve some

aims because contributions cannot be earmarked (IEG 2011a). The evaluation by the World Bank’s

International Evaluation Group (IEG) (2011a: 6) thus concludes: “trust funds are a way to circumvent

the allocation system of the MDBs’ [multilateral development banks] country‐based business

model.”7

Eligibility for IDA funds is based on being under a per capita Gross National Income (GNI)

threshold and on a lack of access to non‐concessionary lending. Replenishments of IDA, and IDA

policies regarding allocations and other issues, are negotiated every third year by donor countries.

Throughout our sample period, IDA resources were allocated according to an explicit rule taking

into account recipient need (as measured by low per capita income) and the quality of economic

policies and governance, where the latter is assumed to increase aid effectiveness (e.g., Burnside and

Dollar 2000, 2004). The World Bank’s “Country Policy and Institutional Assessments” (CPIA)

measure the quality of policies and institutions, and are the most important element in the formula

7 The IEG (2011a) also found that six out of eight donors direct aid resources through trust funds “to issues or countries

of national policy or public interest” and that five out of eight donors use trust funds to influence the World Bank. A UK

official stated that his government supports large global funds in climate change, health and education because of

“impatience with the existing multilateral system” (IEG 2011a: 6).

5

determining allocations for most IDA‐eligible countries. 8 The IDA donors, largely working in

concert with Bank management and staff, determine the content of the CPIA and its weight in the

allocation formula. Morrison (2013) finds that during the Cold War IDA‐eligible countries received

more IDA commitments when they held seats on the World Bank Executive Board. However, he

reports that this effect is absent after 1989, when IDA’s performance‐based allocation system became

more formal and transparent, reducing any discretion in the process. He found no support for the

hypothesis that the CPIA ratings at the core of this allocation system are influenced by the Bank’s

shareholders.

Not all donors, however, are equally supportive of current IDA allocation policies. For

example, some donor countries have called for increasing the weight accorded to need in the

allocation formula, so that countries emerging from conflict may benefit from increased funding (see

Manning 2014).9 Such donors might reduce their contributions to IDA and divert them to supporting

trust funds that complement IDA funding. While donors could also shift these funds to their

respective bilateral aid budgets, they might want some share of it implemented by the Bank. The

IEG (2011a: 6f) evaluation suggests that donors appreciate the World Bank as a trustee because of its

capacity, expertise and strong working relations with governments. If trust fund aid is used in this

way to complement the (cross‐country) allocation of the IDA funds, then trust funds should be more

weakly related than IDA funds to the quality of policies in IDA‐eligible recipient countries – and the

relationship could even be negative.

Trust funds are often used to respond to natural disasters and other humanitarian crises, and

to fund environmental and other projects with significant spillovers across national borders that do

not always conform well to IDA’s country‐based approach. To the extent humanitarian and cross‐

border problems are not correlated with per capita income and policy quality, allocations of trust

funds responding to these problems are likely to be less poverty‐ and policy‐selective than IDA

allocations. For several reasons, therefore, trust funds can potentially undermine IDA’s

“performance‐based allocation” (PBA) system intended (1) to direct more aid to countries where its

development impact is likely to be greater, and (2) to create incentives for policy improvements in

recipient countries.

Our data on disbursements by World Bank trust funds for the 2002‐2012 period allow us to

study whether and how trust fund aid is allocated differently from bilateral and multilateral aid,

particularly with respect to policy‐ and poverty‐selectivity. We analyze trust fund disbursements

using linear regression models that account for time‐invariant recipient characteristics and common

8 The CPIA measures numerous aspects of policies and institutions, grouped into four “clusters” or broad policy areas:

(macro)economic management, structural policies, equity and social inclusion, and public sector management and

institutions. A detailed description can be found at http://siteresources.worldbank.org/PROJECTS/Resources/40940‐

1244163232994/6180403‐1372096800800/CPIAcriteria2012.pdf. 9 Donor countries can and, surprisingly, do unilaterally increase their IDA contributions beyond what is seen as their fair

share (Manning 2014). These additional contributions do not increase the formal voting power of the donor, although

higher voluntary contributions might function as an informal leverage effect for their preferences. It could be that these

additional contributions are popular domestically (Milner 2006). For example, the United Kingdom committed more

than the USA to IDA15 and IDA17 replenishments USA (Manning 2014).

6

year‐specific shocks. Because motives are likely to vary with the type of trust funds (Reinsberg et al.

2015), we consider single‐donor and recipient‐executed trust funds separately, as types where an

individual donor has the greatest control over where and how the funds are used. We also run

separate tests for three sectors in which trust funds are relatively popular with donors – health,

education, and environment ‐ and where the CPIA includes measures of the quality of sectoral

policies.

Several studies (e.g., Annen and Knack 2015; Dollar and Levin 2006) have shown that IDA

disbursements are highly responsive to per capita income and to the quality of policies, as intended

by the IDA performance‐based allocation system. Specifically, they show that disbursements

increase with the quality of policy controlling for per capita income (and population), and that they

decrease with per capita income, controlling for the quality of policy (and population). Aid from

some bilateral donors is also very poverty‐ and/or policy‐selective, but bilateral aid overall is much

less selective than multilateral aid, and IDA is more selective than aid from most other multilaterals

(Knack et al. 2011; Dollar and Levin 2006). Although multilaterals’ allocations are sometimes

influenced by influential donor countries’ political or economic interests (e.g., Barro and Lee 2005;

Kuziemko and Werker 2006; Dreher et al. 2009a, b), those interests matter much more for bilateral

aid allocations (e.g., Alesina and Dollar 2000). Both multilateral and bilateral donors’ aid allocations

have become more selective with respect to the quality of the institutional environment in recipient

countries since the end of the Cold War (Dollar and Levin 2006, Claessens et al. 2009).

Very few analyses have been done of the selectivity of World Bank trust funds. World Bank

(2007) reported that “the distribution of IDA disbursements” among IDA‐eligible countries does not

follow a discernible performance‐based or needs‐based pattern. In fact, it found a negative

correlation with CPIA ratings. However, that report did not explain its methodology: e.g., it is not

clear whether it controlled for per capita income in testing the relationship of trust fund

disbursements with CPIA ratings. A few years later, another World Bank study (Huq 2010) reported

a positive but not very strong unconditional correlation between commitments per capita from

recipient‐executed trust funds (RETF) and the CPIA ratings, which assess recipients’ policy

performance and institutional capacity. An evaluation of the World Bank’s implementation of its

2007 Governance and Anti‐Corruption Strategy reports a positive coefficient on the CPIA rating in

RETF allocation regressions, using aggregate GNI rather than GNI per capita. However, this positive

coefficient is insignificant for RETF disbursements and only marginally significant for RETF

commitments (IEG 2011b: 154). When the sample is limited to IDA‐eligible countries, CPIA is not

significant even for RETF commitments. The analysis is limited to the years 2004‐2010, and it uses

only one of the four “clusters” (on public sector management and institutions) of policy areas in the

CPIA. Wagner (2016: 19) finds that RETF commitments for IDA‐eligible countries during the 2009‐

2013 period are significantly correlated with GNI per capita, population, and CPIA ratings. He

stresses, however, that the collective explanatory power of these three variables is much lower for

RETF allocations than for IDA allocations, suggesting that “trust funds are mainly allocated

according to a different set of criteria.”

7

3. Theoretical Considerations

The World Bank allocates IDA funds among eligible recipients mainly based on its CPIA index that

measures recipients’ policy performance and institutional capacity,10 and on per capita income. This

reliance on the CPIA index reflects donors’ view that resources are more likely to be used

productively in countries with favorable policies and strong public management systems (e.g.,

Burnside and Dollar 2000, 2004). However, donors’ motives when creating World Bank‐managed

trust funds may be very different, and the relationship between CPIA ratings and allocations of trust

funds could be positive or negative. Donors might want to direct more trust funds to higher‐

performing countries for the same reasons as with IDA funds. Alternatively, trust funds might be a

way for some donors to compensate partially for IDA’s performance‐based country allocations, if

they perceive some lower‐performing countries as under‐aided.

Any impact of CPIA ratings – whether positive or negative – should apply more to IDA‐

eligible recipients than to other countries, because the CPIA ratings are used by the Bank only for

IDA allocations. For this reason we run separate regressions for all developing countries and for

IDA‐eligible countries11 only. Because the CPIA does not affect Bank funding to non‐IDA countries,

donors have no reason to compensate for any under‐provision of aid to low‐rated recipients.

When we focus on all recipient countries, we include a dummy for effective IDA eligibility.

The dummy for effective IDA eligibility might influence the probability and size of trust fund

resources with a sign that could go in either direction. On the one hand, IDA recipients might be less

likely than IBRD countries or non‐member states and territories to receive trust fund aid because

they already benefit from IDA resources. On the other hand, bilateral donors might view IDA

countries as those countries in particular need of additional resources (Knack et al. 2014) and where

the Bank has an advantage in expertise, and thus channel more of their “bilateral” aid to those

countries through the Bank in the form of trust funds. In many cases, trust funds are even used to

“co‐finance” IDA projects. If such cases are very common, trust fund aid should be positively

correlated with the IDA dummy, or with net IDA flows, which we control for in some regressions.

We also control for other donor motives in establishing World Bank‐managed trust funds.

Bilateral donors state that they use trust funds to complement their bilateral programming when

their aid agencies do not have sufficient presence or expertise in countries to implement programs

effectively (OECD 2010: 40, 2011: 29), and appreciate the World Bank as a trustee because of its

10 Lack of access to capital markets at market rates of interest is another criterion for IDA eligibility. However, the World

Bank’s assessments of creditworthiness are not disclosed, so we cannot control directly for this variable. We can control

for per capita income and country size (population), which are likely to be important determinants of creditworthiness. 11 This dummy equals one for country‐year observations where there are positive IDA flows or the country is on the list of

eligible IDA recipients. We apply these two criteria which are not perfectly congruent. Some countries might not want to

borrow from the IDA while some countries still borrow because they are just above the eligibility cutoff (blend countries).

Note that according to Huq (2010, footnote 9), financial support from IDA is not available to Sudan because of outstanding

arrears. Therefore, we set the IDA eligibility dummy for Sudan to zero.

8

capacity, expertise and strong working relations with governments (IEG 2011a: 6f). Specifically,

donors might provide funding through the Bank rather than directly in circumstances where they

do not want to be present in recipient countries themselves. For example, donors might want to

delegate project implementation to the Bank in fragile countries because it allows diffusing

accountability about aid effectiveness, because sending bilateral staff is politically sensitive, or

because donors want to act in concert through the Bank to avoid harmful fragmentation in aid

activities. We test for the relationship between trust fund and bilateral aid by directly including

bilateral development assistance in some regressions.

Rhetoric by donors and the World Bank also suggests that trust funds are a useful instrument

in fragile contexts. We expect fragile countries to be more likely to receive aid from trust funds rather

than from IDA. On the one hand, the World Bank might be restricted by its legal mandate when

governments in fragile contexts change repeatedly, making it virtually impossible to negotiate

programs. Fragile states also have low CPIA scores by definition, limiting the IDA resources

available for this country mechanically due to the allocation rule.12 On the other hand, some bilateral

donors may view the IDA performance‐based allocation system as under‐aiding fragile countries

confronted with challenging situations. Moreover, some donors might be more concerned than other

donors with security, refugee and other problems associated with specific fragile‐ and conflict‐

affected states. Such donors might be geographically proximate to the fragile situation, or have high

reputation or economic stakes in the fragile country, such as former colonial powers. World Bank

(2007) reports that RETF disbursements to fragile states increased from $130 million to $679 million

between fiscal years 2002 and 2006.

Donors also use trust funds to fill “gaps in the multilateral system,” in particular to provide

a rapid response to “emergencies such as natural disasters, disease outbreaks, and the end of armed

conflict, where donors want to coordinate their bilateral aid and where the MDBs do not grant

resources to engage on a sufficient scale” (IEG 2011a: 5). We test these motives by including dummy

variables for disaster‐affected, post‐conflict, and fragile states.

Given the multitude of motives for using trust funds, we use multivariate statistical analysis

that allows assessing the relative importance of these simultaneous motives while keeping time‐

invariant characteristics and year‐specific shocks constant.

12 A small number of post‐conflict countries receive supplemental IDA funding for a limited number of years.

9

4. Data

Our dependent variable is (logged) disbursements of trust funds to recipient countries based on the

financial accounting tables of the World Bank.13 In the disbursements data, recipient countries are

almost always indicated but donors are unknown, so that we cannot discriminate between the

sources of funds. 14 The unit of analysis thus is the recipient country‐year, rather than donor‐

recipient‐year. For purposes of comparing the overall aid selectivity of World Bank trust funds to

IDA and to other bilateral aid, this is the appropriate unit of analysis, because it implicitly weights

larger trust fund donors more heavily in the analysis. With donor‐recipient‐year observations, each

donor‐recipient pair would receive an equal weight. If selectivity of trust fund aid differed

systematically for large and small donors, selectivity coefficients could be a misleading indicator of

overall selectivity.15

The structure of the disbursements data is depicted in Figure 2. The figure indicates that trust

funds may differ along three dimensions: by the number of contributors (Figure 3), the country‐

specific, regional or global mandate, and by their execution type (Figure 4). We analyze a subset of

the universe of World Bank funds, IBRD/IDA trust funds, which are most numerous. The Bank

makes a technical distinction among three types of trust funds. The Bank only provides financial

services to Financial Intermediary Funds (FIFs).16 The number of trust funds of the International

Finance Corporation (IFC) is relatively minor and has different objectives than IBRD/IDA trust

funds. IBRD/IDA trust funds, the focus of this analysis, are classified further by recipient‐executed

trust funds (RETFs), which are implemented by a third party but supervised by the Bank, and Bank‐

executed trust funds (BETFs), which support the Bank’s work directly. While RETFs are similar to

the IDA or IBRD in terms of being disbursed to recipient countries, BETFs are more similar to Bank

administrative expenses, and often but not always finance Bank activities that are not country‐

specific and thus not relevant for our research question (Figure 4).

In testing for poverty‐ and policy‐selectivity of trust funds and other aid, we control for the

natural logarithm of population. The IDA allocation formula provides more aid to larger countries,

but with an elasticity of less than one. Studies typically show a “small country bias” for IDA

disbursements – i.e., aid increases less than proportionately with population – but this bias is smaller

than for aid from most other donors (e.g., Annen and Knack 2015).

Due to missing and low‐quality data on actual poverty rates (e.g., the percentage share of the

population living on less than $2 per day), income per capita is used as a proxy for poverty in the

IDA formula. In the tests we report below we follow the IDA allocation formula specifically by using

13 We use the data from Reinsberg et al. (2015) as obtained from the Bank’s CFGP. 14 The by far most important contributors to trust funds are DAC countries, accounting for 80 percent of contributions

over the period of study. Non‐DAC donor countries, private companies, NGOs and multilateral organizations are

relatively minor contributors (see also Eichenauer 2015). 15 In Wagner (2016), CPIA ratings are significant in regressions based on recipient ‐year observations, but not in

regressions based on donor‐recipient‐year observations. 16 Our data set does not contain the information on country allocation by FIFs, which have their own governance and

disbursement systems (see Reinsberg et al. (2015).

10

GNI per capita, smoothed for exchange rate changes using the Atlas method, rather than GDP per

capita, which is used more commonly in the aid allocation literature. The World Bank also uses GNI

per capita in its lending classifications (including eligibility for IDA) and its income classifications.

Unsurprisingly, poverty selectivity of World Bank‐administered aid (IDA or trust funds) tends to be

somewhat weaker if we use GDP per capita in our regressions instead of the measure the World

Bank uses in its own allocation policies, GNI per capita. The choice of GNI or GDP makes little

difference, however, for most other results, including the policy selectivity of aid.17 Data for both

income measures and for population come from the World Development Indicators (WDI 2015).

Aggregated and partially‐disaggregated CPIA scores for IDA‐eligible countries are publicly

available from the WDI (2015) since 2005, and range from a minimum score of 1 (lowest quality

policies) to a maximum of 6 (highest quality). For non‐IDA countries and prior years, scores are not

publicly available, but were obtained with the necessary permissions by one of the authors from

internal World Bank databases. In contrast to Wagner (2016), we are therefore able to include

observations dating to 2002, and for non‐IDA as well as IDA countries. In a few cases, CPIA scores

are not assigned after countries fall into arrears with the World Bank and are no longer classified as

active borrowers. This typically is the case when the quality of policies and governance are very low,

as indicated by their CPIA scores for years when they were active borrowers. In these cases, we

replace the missing score with the lowest score received by any other country in this year, but results

are not sensitive to dropping these observations instead.

Our main tests do not include other control variables, because our interest is mostly in

comparing selectivity across aid sources. We control in some tests for a fragile state dummy, because

some donors in recent years have given them favored status in allocating aid. A very small share of

IDA funds are set aside for a handful of “post‐conflict” countries that overlap considerably with

fragile state status. We create a dummy variable for fragile country status from the official World

Bank list of fragile and conflict situations, first produced in 2006 and updated annually. When we

include the fragile dummy, we limit the sample to the fiscal years 2006‐2012, where the dummy

equals one in those country‐years in which a country was on the official lists of fragile situations.18

We create a separate dummy for countries eligible for special “post‐conflict” funding (World Bank

2014b). In some specifications, we also control for (logged) bilateral aid and development aid

received from official donors through channels other than trust funds and IDA using OECD/DAC

data (2015) and World Bank data for (logged) IDA flows. If other aid flows are to some extent policy‐

and poverty‐selective, these aid flows are collinear with our measures of policy quality and poverty

for which the coefficients are then imprecisely estimated.

In robustness tests, we control for several donor‐interest variables, following the literature

on allocation of bilateral aid (e.g., Alesina and Dollar 2000). First, we measure geopolitical interests

17 Full results using GDP per capita are available from the authors upon request. In reporting results below we note two

instances in which significant results for one variable become insignificant when GDP is used instead of GNI. 18 The absence of a list prior to 2006 likely reflects the lack of any influential view among donors that “fragility” was an

important concept or that fragile states should be treated differently.

11

by the alignment of votes between recipient and donors in the United Nations General Assembly

(UNGA) using data from Strezhnev and Voeten (2009). Annual measures of voting alignment range

from 0 to 1 with higher values implying higher similarity and are calculated as in Kilby (2013) using

all UN votes. We calculate the mean alignment of the largest three sovereign trust fund donors, the

United Kingdom, the United States and the Netherlands, which we label as “the G3.”19 We exclude

the European Commission, the second largest donor, because its preference aggregation process is

more complex (e.g., Reinsberg et al. 2014). Second, we add a dummy for a colonial relationship

between a G3 donor and the recipient country. Third, commercial interests are measured by total G3

exports, using data from the IMF (2015). A thorough test of the importance of these donor‐interest

variables in aid allocation decisions would require a different data set, with donor‐recipient‐year

observations. In our analysis, we merely use these as control variables.

Table 2 displays the descriptive statistics for the dependent and control variables in our

sample. The average recipient country receives 1.8 USD trust fund aid per capita, a relative minor

inflow compared to average per capita aid of 6.7 USD from IDA20 and 67 USD from other sources.21

However, trust fund volumes have increased over the sample period and represent significant

inflows for some recipient countries as indicated by the maximum value of more than 82 USD trust

fund aid per capita. Average GNI per capita is 2,744 USD, which the Bank classified as a lower

middle income country in the fiscal year 2012. The mean of the aggregate CPIA rating is just below

the median value of 3.5, and the maximum value is almost 5. Recipient countries’ education, health

and environmental policy ratings range between 1 and 6 with an average value similar to the

aggregate CPIA score. With regard to the political and economic variables, the data show that

average voting affinity with the largest donors is slightly higher than the median value of 0.5. Total

exports per capita from the G3 trust fund donors (UK, USA, and the Netherlands) vary substantially

around the mean value of 202 USD per capita. The table further shows that one third of recipient

countries in our sample are former G3 colonies, fragile countries make up 17 percent, and IDA‐

eligible countries 65 percent, of observations.

5. Estimation Method and Results

This section compares the selectivity of trust fund aid to IDA and bilateral aid. We use panel models

with year‐ and country‐fixed effects without a selection stage, because almost all recipient countries

receive positive trust fund disbursements at least once during our sample period. Our sample

includes all countries that the OECD’s Development Assistance Committee (DAC) considered

developing countries in a given year.

As is common in the aid allocation literature, our baseline regression model looks as follows:

19 In the fiscal years 2002‐2012, these donors contributed 4.08, 2.47, and 2.04 billion constant USD respectively. The

European Commission contributed 2.47 while Canada as the fifth largest donor contributed 1.12 billion USD. 20 This figure is based on the sample of all developing countries. Among IDA eligible countries, the figure is 10.3 USD

per capita. 21 These high per capita figures are driven by island states.

12

ln(disbursements)i,t = α + β CPIAi,t‐1 + γ ln(populationi,t‐1) + δ ln(GNI p.C.i,t‐1) + λ’Xi,t‐k + αi + σt + εi,t

where i refers to the recipient country and t to the Bank’s fiscal year. Errors εi,t are robust to

heteroscedasticity and clustered at the recipient country level and we include time‐fixed effects σt

and, in most regressions, recipient fixed effects αi.

The inclusion of recipient fixed effects controls for any time‐invariant recipient characteristics

that may influence aid allocations: estimates of poverty‐ and policy‐selectivity are then informed

only by changes over time within countries in aid, GNI per capita, and quality of policy.22 However,

we also report results from pooled OLS (with standard errors clustered by recipient), in which

estimates are informed primarily by cross‐country rather than over‐time variation in the data. The

CPIA ratings are designed to compare the quality of policies across countries at a point in time using

all relevant information at that time. Changes in ratings from one year to the next sometimes reflect

newly‐available information rather than actual changes in the quality of policies and institutions,

and ratings from prior years are not “corrected.” Over‐time variations in the data therefore are likely

to incorporate substantially more measurement error than in the cross‐sectional variation.

Reassuringly, the fixed effects estimates (which minimize omitted variable bias) reported below are

largely consistent with the pooled OLS estimates (which are likely based on less noisy data).

Table 3 shows results with recipient‐ and year‐fixed effects for all IDA countries (columns 1‐

3) and all developing countries (columns 4‐6). Trust funds and IDA provide significantly higher aid

to poorer and better‐governed (i.e. higher CPIA) countries (columns 3 and 1 respectively). Trust

fund allocations are thus much more similar to IDA than to other ODA in terms of poverty and

policy selectivity. 23 Coefficients on (log of) GNI per capita (and on log of population) are

interpretable as elasticities, so a 1% increase in per capita GNI reduces IDA flows by 1.9% (column

1) and TF aid by 2.5% (column 3). A 1‐point increase in the CPIA rating (measured on a 1 to 6 scale)

is associated with a nearly 20‐fold increase in IDA flows ( = e3.031- 1) and 7-fold increase in TF aid (

= e2.088- 1). Note, however, that a 1‐point increase in CPIA is very large in relative terms, equal to

nearly two standard deviations.

Even for a sample that includes non‐IDA countries (column 6), trust funds are very policy‐

selective as well as poverty‐selective. As shown in Table 4, this holds true in OLS regressions with

year‐fixed effects that allow for the inclusion of time‐invariant variables. The main difference is that

the estimated effects of country size are more consistent and significant. The coefficients on

population for other aid types in Table 4 reflects the bias in favor of small‐countries (in per capita

terms) frequently noted in the literature, but this bias is nearly absent for trust fund aid, where the

coefficient on population is even slightly above one (see Fleck and Kilby 2010). When we include a

dummy for fragile countries and limit the sample to the post‐2005 years accordingly, we do not find

that these countries receive more trust fund aid (Table 5, column 2). Moreover, the inclusion of this

22 Hausman tests show that fixed effects are preferred over random effects. 23 There is no statistical difference between the coefficients on the selectivity variables for IDA and TF aid, while

coefficients are significantly different between trust fund aid and other aid flows (p‐value=0.002 for GNI p.c. and p‐

value=0.023 for the CPIA score). We run SUR models to test the equality of coefficients using the suest command in Stata.

13

dummy has only trivial effects on coefficients for the selectivity variables (comparing columns 1 and

2).

Potentially, allocating aid partly for commercial and political motives could weaken the

policy‐ and poverty‐selectivity of trust funds. If so, then when we control for those factors the

absolute value of the selectivity coefficients should increase, i.e. “conditional” selectivity should be

stronger than “unconditional” selectivity. We include measures for commercial and political

motives in Table 5 (column 3), but find the coefficients on the selectivity variables to change little

compared to Table 3 (column 6). The commercial and political variables ‐ UNGA voting alignment

with the G3, and (log of) total G3 exports ‐ are significant, with the expected positive signs.24 In

column 4 of Table 5 we drop the country‐fixed effects, so we are able to add a second political interest

variable, a dummy for former colonies of G3 donors. The colonial heritage dummy is not significant

and its inclusion increases the size and significance level of the selectivity coefficients (comparing

Table 5, column 4 and Table 4, column 6). The last column in Table 5 includes other aid flows and

IDA aid, where the latter coefficient is positive and marginally significant.25 This finding reinforces

the idea that trust fund allocations look a lot more like IDA’s than like other aid. But, more

importantly, it is showing that donors’ trust fund allocations are not just guided by IDA flows,

including co‐financing of IDA projects: even controlling for IDA flows, their trust fund allocations

are separately influenced by IDA policies, namely poverty and policy selectivity.

Donor countries might prefer RETFs, which are under more direct control of donors, over

trust funds more generally. In particular, some BETF allocations by country are determined by staff.

Therefore, RETFs could be less poverty‐ and policy ‐ selective, and guided by political and

commercial objective to a larger extent. Column 2 of Table 6 replicates the method and specification

of Table 5, column 3, but for RETFs instead of for all TFs. In this fixed effects test, GNI per capita is

not significant, and its coefficient is reduced in absolute value compared to the case of all TFs. The

political and commercial variables also are not significant. In the OLS test, per capita income is also

insignificant while political alignment is associated with significantly larger RETF allocations (Table

6, column 4). In the OLS tests (comparing column 3 to column 4, the coefficient for CPIA increases

and becomes significant by controlling for political and commercial interests.

Single‐donor trust funds could also be used by donors to support strategic interests, while

multi‐donor trust funds (MDTFs) should look more like IDA: other things equal, as an MDTF

includes more donors, it will look more like the full set of IDA donors. Compared to an SDTF donor,

donors in MDTFs should therefore have a stronger collective interest in selectivity while

disagreement about countries to benefit from favors should limit the role of political and commercial

factors in determining aid allocations. Columns 5‐8 of Table 6 replicate columns 1‐4, but for SDTFs

instead of RETFs. We find some support for the hypothesis that political interests matter more for

SDTFs, as coefficients on voting alignment are somewhat larger in columns 6 and 8 than in columns

24 In column 3, the coefficient falls from 6.1 to 4.2 and becomes insignificant if we measure income using GDP instead of

GNI per capita. 25 For the IDA sample, IDA flows are significant in the fixed and random effects models.

14

3 and 4, respectively, of Table 5.26 These coefficients are not significantly larger, however, and

selectivity for policy and poverty is still strong. In sum, we find that, despite being guided by

political and commercial motives to a somewhat larger extent than aggregate trust fund aid, RETFs

remain policy‐selective while SDTFs remain both policy‐ and poverty‐selective.

Table 7 tests whether donors use trust funds to provide extra funding for post‐conflict and

disaster‐affected countries in a coordinated manner. We find no systematic evidence that these

motives explain trust fund allocations. Neither a post‐conflict dummy variable nor a measure of the

number of people affected by a disaster is significantly related to TF aid, in either fixed effects or

pooled OLS regressions, for either the IDA sample or the full developing‐country sample.

In Table 8, we disaggregate the data to analyze three specific sectors – education, health, and

environment – that are popular among trust fund donors (Figure 5), and for which sector‐specific

indicators of the quality of policies (CPIA sub‐ratings) are produced. For this analysis, we use the

relevant sector‐specific IDA disbursements, as well as the sector‐specific CPIA sub‐ratings. Most

projects in the sample have objectives in multiple sectors, and we assign projects to the sector which

has the largest sectoral share. We focus on the two sectors for which we have a respective CPIA sub‐

score, namely education and health. For environment, we use the theme code and the corresponding

CPIA rating. 27 We find health and education aid to be poverty‐ and policy‐selective in OLS

regressions with time dummies (Table 8, columns 1 and 3). While education aid is responsive also

to policy improvements over time within countries (column 2), health aid is not (column 4). In

contrast, environmental aid is responsive neither to policy improvements over time within countries

(i.e., in the fixed effects regression reported in column 6) nor between countries (column 5).

In summary, we find consistent evidence that the cross‐country allocation of aggregate trust

fund aid is poverty‐ and policy‐ selective and allocated more similarly to IDA than to other aid.

Controlling for population, per capita income, and the quality of policies, we find no evidence that

fragile, post‐conflict, or disaster‐affected countries receive more trust fund aid. Sector‐specific

analyses, using sector‐specific policy indicators, show that trust fund aid in education and (to a lesser

extent) health is sensitive to population, income and the quality of policies, but allocations of

environmental trust funds are unrelated to the quality of environmental policies.

26 In column 6, the coefficient falls from 7.4 to 6.2 and becomes insignificant if we measure income using GDP instead of

GNI per capita. 27 Criteria for the education, health and environment policy ratings in the CPIA can be found at

http://siteresources.worldbank.org/PROJECTS/Resources/40940‐1244163232994/6180403‐

1372096800800/CPIAcriteria2012.pdf.

15

6. Conclusion

This paper provides the most comprehensive analysis to date of the cross‐country allocation of trust

funds, a relatively new aid channel that can be described as a hybrid between bilateral and

multilateral aid. Donor rhetoric suggests that some World Bank‐administered trust funds are

intended to support countries neglected by IDA for legal, political, and other reasons. Potentially,

therefore, the cross‐country allocation of these trust funds may compensate for the performance‐

based allocation of IDA funds, “diluting” the poverty‐ and policy‐selectivity of total World Bank‐

administered aid. We test whether donors’ allocations of trust funds tend to favor recipients with

lower per capita incomes and more favorable policy environments, as measured by the Bank’s CPIA.

Results show that World Bank trust fund allocations are similar to IDA, and dissimilar to other

bilateral aid (summed over all DAC donors) in terms of their policy‐ and poverty‐selectivity.

The allocation of recipient‐executed and single‐donor trust funds, which are under closer

control of donor countries, appears to be more strongly motivated by strategic interests of donor

countries than trust fund aid in general. However, a more complete investigation of the role of donor

interest variables would require a different data set and different unit of analysis (donor‐recipient

pairs) than we use in this paper.28 We also explore trust fund disbursements for three specific sectors,

and find evidence that health and education aid, but not environmental aid, are policy and poverty

selective across countries.

Overall, the evidence indicates that multi‐bi funds administered by the World Bank do not

undermine IDA’s allocation criteria, even if donors’ contributions to them partially crowd out their

IDA contributions. There is very little evidence regarding the latter issue of crowding out, and we

leave for future research the question of whether the increase in trust funds has affected core

contributions to IDA, and their potential effects on other aspects of World Bank operations.29 The

question of the additionality of earmarked funds at the aggregate level remains open as donors

might just reshuffle funds away from multilateral organizations they perceive as less effective.

Finally, as experience with trust fund aid accumulates, another important research question will be

its effectiveness, relative to core multilateral and traditional bilateral aid, in contributing to growth

and human development outcomes.

28 Factors related to donors’ and recipients’ domestic political economies are investigated by Dietrich (forthcoming) and

Eichenauer and Reinsberg (2016). 29 One review of World Bank non‐lending technical assistance projects rated the quality of fully‐trust funded projects

lower than projects that were wholly or partially Bank‐financed, and concluded that the difference was attributable in

part to less attention by management and staff to trust funded projects (see IEG 2011a: ch. 3).

16

Bibliography

Alesina, Alberto and David Dollar (2000). Who gives foreign aid to whom and why? Journal of

Economic Growth 5(1): 33‐63.

Annen, Kurt and Stephen Knack (2015). On the delegation of aid implementation to multilateral

agencies. World Bank Policy Research Working Paper No. 7455.

Barakat, Sultan (2009). The failed promise of multi‐donor trust funds: Aid financing as an

impediment to effective state‐building in post‐conflict contexts. Policy Studies 30(2): 107‐126.

Barakat, Sultan, Kathryn Rzeszut, and Nick Martin (2012). What is the track record of multi‐donor

trust funds in improving aid effectiveness? An assessment of the available evidence.

University of London, EPPI‐Centre Report No. 2005, Institute of Education.

Barro, Robert and Jong‐Wha Lee (2005). IMF Programs: Who is chosen and what are the effects?

Journal of Monetary Economics 52(7): 1245‐1269.

Burnside, Craig and David Dollar (2000). Aid, policies and growth. American Economic Review 90(4):

847‐868.

Burnside, Craig and David Dollar (2004). Aid, policies and growth: Reply. American Economic Review

94(3): 781 – 784.

Claessens Stijn, Danny Cassimon, and Bjorn Van Campenhout (2009). Evidence on changes in aid

allocation criteria. World Bank Economic Review 23(2): 185‐208.

Dietrich, Simon (forthcoming). Donor political economies and the pursuit of aid effectiveness.

International Organization.

Dollar, David and Victoria Levin (2006). The increasing selectivity of foreign aid: 1984‐2003. World

Development 34(12): 2034‐2046.

Dreher, Axel and Sarah Langlotz (2015). Aid and growth. New evidence using an excludable

instrument. CEPR Discussion Paper No. 10811.

Dreher, Axel, Jan‐Egbert Sturm and James R. Vreeland (2009a). Development aid and international

politics: Does membership on the UN Security Council influence World Bank decisions?

Journal of Development Economics 88(1): 1‐18.

Dreher, Axel, Jan‐Egbert Sturm and James R. Vreeland (2009b). Global horse trading: IMF loans for

votes in the United Nations Security Council. European Economic Review 53: 742‐757.

Eichenauer, Vera Z. (2015). Trust funds: DAC donors contribute, most non‐DAC donors don’t. First

Tranche Blog by AidData. February 20.

Eichenauer, Vera Z. and Bernhard Reinsberg (2016). What determines earmarked funding to

international development organizations? Evidence from the new multi‐bi dataset. Presented

at the Political Economy of International Organization Conference, Salt Lake City, January

2016.

17

Eichenauer, Vera Z. and Hug, Simon (2014). The politics of special purpose trust funds. Presented at

the Political Economy of International Organization Conference, Princeton, February 2014.

EM‐DAT (2014). The international disaster database. Centre for Research on the Epidemiology of

Disasters. Accessed on June 16, 2014: http://www.emdat.be/database

Fleck, Robert K. and Christopher Kilby (2010). Changing aid regimes? U.S. foreign aid from the Cold

War to the War on Terror. Journal of Development Economics 91: 185‐197.

Huq, Wahida (2010). Analysis of Recipient Executed Trust Funds. CFP Working Paper Series 5.

Concessionary Finance and Global Partnerships Vice Presidency, World Bank, Washington

DC.

IEG (2011a). An evaluation of the World Bankʹs trust fund portfolio: Trust Fund support for

development. Washington DC: Independent Evaluation Group.

IEG (2011b). World Bank country‐level engagement on governance and anticorruption: An

evaluation of the 2007 Strategy and Implementation Plan. Washington DC: Independent

Evaluation Group.

IMF (2015). Direction of trade statistics (DOTS). Washington, DC: International Monetary Fund.

Download on May 20.

Kilby, Christopher (2013). The political economy of project preparation: An empirical analysis of

World Bank projects. Journal of Development Economics 105: 211‐225.

Knack, Stephen, Colin Xu and Ben Zou (2014). Interactions among donors’ aid allocations: Evidence

from an exogenous World Bank income threshold. World Bank Policy Research Working

Paper No. 7039.

Knack, Stephen, Halsey Rogers and Nick Eubank (2011). Aid quality and donor rankings. World

Development 39(19): 1907‐1917.

Kuziemko, Iliana and Erik Werker (2006). How much is a seat on the Security Council worth?

Foreign aid and bribery at the United Nations. Journal of Political Economy 114(5): 905–930.

Manning, Richard (2014). The multilateral aid system: An assessment following the major

replenishments of 2013. WIDER Working Paper No. 110.

Milner, Helene (2006). Why multilateralism? Foreign aid and domestic principal‐agent problems. In

D. G. Hawkins et al. (Eds.), Delegation and agency in international organizations. New York:

Cambridge University Press.

Morrison, Kevin M. (2013). Membership no longer has its privileges: The declining informal

influence of Board members on IDA lending. The Review of International Organizations 8(2):

291‐312.

OECD (2015). Official Development Assistance data. Paris: Organisation for Economic Co‐operation

and Development.

18

OECD (2011). 2011 DAC report on multilateral aid. Paris: Organisation for Economic Co‐operation

and Development.

OECD (2010). 2010 DAC report on multilateral aid. Paris: Organisation for Economic Co‐operation

and Development.

Reinsberg, Bernhard (2015). The implications of multi‐bi financing on international development

organizations: The example of the World Bank. In Mahn, T., Negre, M., and Klingebiel, S.

(Eds.) Fragmentation or Pluralism? The Future of Development Cooperation Revisited. Basingstoke:

Palgrave McMillan.

Reinsberg, Bernhard, Katharina Michaelowa, and Vera Z. Eichenauer (2015). The rise of multi‐bi aid

and the proliferation of trust funds. In Arvin, B. M. & Lew, B. (Eds.) Handbook on the Economics

of Foreign Aid. Cheltenham: Edward Elgar.

Reinsberg, Bernhard, Katharina Michaelowa and Stephen Knack (2015). Which donors, which

funds? The choice of trust funds by bilateral donors at the World Bank. World Bank Policy

Research Working Paper 7441.

Michaelowa Katharina, Bernhard Reinsberg and Christina Schneider (2013). When international

organizations delegate: The politics of earmarking European Union aid in multilateral aid

institutions. Unpublished manuscript.

Strezhnev, Anton and Erik Voeten (2009). United Nations General Assembly voting data V7.

Downloaded on August 17, 2014 from: http://hdl.handle.net/1902.1/12379.World Bank

(2014a). Fragile situations lists. Washington DC: World Bank. Accessed on August 20, 2014:

http://go.worldbank.org/BNFOS8V3S0.

OECD/DAC (2015). Creditor Reporting System. Paris: Organization for Economic Cooperation and

Development.

Wagner, Laurent (2016). ʺHow do Earmarked Funds Change the Geographical Allocation of

Multilateral Assistance?ʺ Ferdi Working Paper P150.

\World Bank (2014a). List of Fragile Countries. Washington DC: World Bank Development Finance

department memo.World Bank (2014). Countries eligible for IDA post‐conflict allocation.

Washington DC: World Bank Development Finance department memo, updated April 2014.

World Bank (2007). A management framework for World Bank administered trust funds. Washington, DC:

World Bank. http://documents.worldbank.org/curated/en/2007/09/8618148/management‐

framework‐world‐bank‐administered‐trust‐funds.

WDI (2015). World Development Indicators. Washington DC: World Bank.

19

Figure 1: Disbursements by IDA and World Bank trust funds

Source: World Bank trust fund data, 2002‐2012. Disbursements by trust funds and IDA excluding

disbursements by FIFs and IFC trust funds.

Figure 2: Data structure and types of trust funds

20

Figure 3: Disbursements by single‐ and multi‐donor trust funds



Figure 4: Disbursements by execution type



21

Figure 5: Trust fund disbursements by sectors



22

Table 1: Descriptive statistics

Sample mean Standard

deviation

Minimum Maximum

Aid flows per capita

Trust fund aid per capita 1.84 6.68 0.00 82.36

IDA aid per capita 6.73 11.50 0.00 138.62

Other aid per capita 66.70 180.78 0.00 2,469.26

Performance variables

GNI per capita, constant USD 2,744 2,743 131 14,532

Population in millions 42 156 0 1,344

Agg. CPIA score 3.43 0.57 1.40 4.91

CPIA, Education 3.54 0.68 1.00 6.00

CPIA, Health 3.50 0.67 1.00 6.00

CPIA, Environment 3.31 0.65 1.00 6.00

Political economy variables

Voting affinity in UNGA (G3) 0.56 0.15 0.21 0.92

Total exports per capita, USD (G3) 203 432 1 4,818

Former colonies (G3) 0.35 0.48 0.00 1.00

Other control variables

Total disaster affected (3‐year total)* 2,048 14,209 0 185,932

Fragile countries 0.02 0.16 0.00 1.00

Post conflict situation 0.17 0.38 0.00 1.00

IDA eligible countries 0.65 0.48 0.00 1.00

Note: * In thousands

23

Table 2: Variable sources and definitions

Variable Definition Original Source

Trust fund disbursements Disbursements by World Bank trust funds (logged, FY, *) Trust Funds and Partnerships Department at the World

Bank

Bilateral aid disbursements Official Development Aid disbursements (logged, *) DAC, Table DAC2a ODA Disbursements

GNI per capita Gross national income, Atlas method (logged,*) World Development Indicators (2015)

GDP per capita Gross domestic product per capita in PPP (logged, *) World Development Indicators (2015)

Population Population of recipient and donor countries (logged) World Development Indicators (2015)

IDA aid ODA by the International Development Association

(IDA) (logged, FY, *) World Bank data from S. Knack

IDA dummy IDA eligible countries and IDA recipient countries (incl.

blend countries, FY) S. Knack and IDA aid data

G3 Colonial Heritage Former colony of a G3 donor (US, UK, Netherlands) CIA Factbook and other sources

G3 UNGA voting alignment Voting alignment with G3 donors in United Nations

General Assembly, calculated as Kilby (2013)

Streznev and Voeten (2009), important years from

Christopher Kilby

Exports Total value of exports from G3 donors to recipient

country (logged,*) IMF DOTS (2015)

CPIA scores Aggregated and sector‐disaggregated Country Policy and

Institutional Assessment scores (1‐6)

Aggregated CPIA since 2006 from World Development

Indicators (2015); Other scores from S. Knack

Post conflict situation Dummy equals one for country‐years eligible for IDA

special allocations for post‐conflict countries World Bank (2014b)

Fragile state dummy Dummy equals one for country‐years classified as fragile

situation by the World Bank Own coding based on World Bank (2014a)

Disaster affected (total)

Number of disaster‐affected people in the recipient

country, moving 3‐year total (contemporaneous, once and

twice lagged, logged)

EM‐DAT: The OFDA/CRED International Disaster

Database

DAC recipients Historical list of DAC recipients (1997‐2013) Dreher and Langlotz (2015)

Notes: Data in calendar years if not marked as fiscal year (FY). * Values are transformed to 2011 constant USD using the GDP‐Deflator from WDI (2015).

24

Table 3: Comparison of poverty and policy selectivity across TFs, IDA, and other aid, fixed effects

1 2 3 4 5 6

GNI p.c. (ln) ‐1.899** 0.764 ‐2.518*** ‐1.028** 1.396 ‐3.430***

[0.828] [0.813] [0.815] [0.508] [0.962] [0.783]

CPIA score 3.031*** ‐0.33 2.088*** 1.953*** ‐0.827 2.596***

[1.013] [0.891] [0.761] [0.717] [0.856] [0.794]

Population (ln) 3.082 0.018 2.864 3.034 ‐5.64 2.23

[2.523] [3.302] [3.206] [1.975] [6.261] [2.737]

IDA eligible ‐ ‐ ‐ 15.538*** ‐2.52 0.656

‐ ‐ ‐ [1.506] [2.556] [1.031]

Dependent variable IDA Other TF IDA Other TF

Sample IDA IDA IDA All All All

Estimation FE FE FE FE FE FE

Adjusted R‐squared 0.075 0.002 0.189 0.316 0.015 0.166

# Observations 900 900 900 1,375 1,375 1,375

# Countries 89 89 89 133 133 133

Notes: The dependent variables is the natural logarithm of the aid type noted in the bottom of

the respective column and covers the Bankʹs fiscal years 2002‐2012. All regressions include

dummies for the World Bank’s fiscal year. Robust standard errors in parentheses clustered at

the recipient country level. * p<0.10, ** p<0.05, *** p<0.01

Table 4: Comparison of poverty and policy selectivity across TF aid, IDA, and other aid, OLS

1 2 3 4 5 6

GNI p.c. (ln) ‐1.571*** ‐0.735* ‐0.647 ‐1.238*** ‐0.817* ‐0.783**

[0.512] [0.374] [0.403] [0.362] [0.436] [0.356]

CPIA score 5.337*** ‐0.502 1.809** 2.601*** ‐0.424 0.924

[1.330] [0.445] [0.714] [0.849] [0.475] [0.624]

Population (ln) 0.771*** 0.511*** 1.088*** 0.495** 0.353* 1.156***

[0.262] [0.150] [0.204] [0.200] [0.182] [0.155]

IDA eligible ‐ ‐ ‐ 15.011*** 0 1.137

‐ ‐ ‐ [0.777] [0.840] [0.741]

Dependent variable IDA Other TF IDA Other TF

Sample IDA IDA IDA All All All

Estimation OLS OLS OLS OLS OLS OLS


# Observations 900 900 900 1,375 1,375 1,375

# Countries 89 89 89 133 133 133

Notes: The dependent variables is the natural logarithm of the aid type noted in the bottom of the

respective column and covers the Bankʹs fiscal years 2002‐2012. All regressions include dummies

for the World Bank’s fiscal year. Robust standard errors in parentheses clustered at the recipient

country level. * p<0.10, ** p<0.05, *** p<0.01

25

Table 5: Robustness checks

1 2 3 4 5

GNI p.c. (ln) ‐4.955*** ‐4.961*** ‐3.759*** ‐0.839** 3.274***

[1.386] [1.389] [0.806] [0.373] [0.801]

CPIA score 3.594** 3.541** 2.380*** 1.058** 2.324***

[1.515] [1.571] [0.804] [0.473] [0.840]

IDA eligible 0.841 0.84 0.781 1.866*** ‐1.459

[1.070] [1.070] [1.163] [0.676] [1.475]

Population (ln) 4.169 4.035 1.234 0.978*** 1.751

[4.414] [4.395] [2.735] [0.260] [2.739]

Fragile situation ‐ ‐0.154 ‐ ‐ ‐

‐ [0.507] ‐ ‐ ‐

G3 UNGA‐Alignment ‐ ‐ 6.149* 8.439*** ‐

‐ ‐ [3.581] [2.251] ‐

Total G3 exports (ln) ‐ ‐ 0.622** 0.13 ‐

‐ ‐ [0.304] [0.214] ‐

G3 colonial heritage ‐ ‐ ‐ ‐0.692 ‐

‐ ‐ ‐ [0.456] ‐

Other aid (ln) ‐ ‐ ‐ ‐ ‐0.013

‐ ‐ ‐ ‐ [0.025]

IDA aid (ln) ‐ ‐ ‐ ‐ 0.134*

‐ ‐ ‐ ‐ [0.080]

Dependent variable TF TF TF TF TF

Sample >2005 >2005 All All All

Estimation FE FE FE OLS FE

Adjusted R‐squared 0.164 0.164 0.177 0.391 0.173

# Observations 887 887 1,332 1,332 1,375

# Countries 133 133 128 128 133

Notes: The dependent variables is the natural logarithm of trust fund aid and covers

the Bankʹs fiscal years 2002‐2012. All regressions include dummies for the World

Bank’s fiscal year. Robust standard errors in parentheses clustered at the recipient

country level. * p<0.10, ** p<0.05, *** p<0.01

26

Table 6: Recipient‐executed and single‐donor trust funds

1 2 3 4 5 6 7 8

GNI p.c. (ln) ‐1.559 ‐1.757 ‐0.517 ‐1.039

‐

3.188***

‐

3.080***

‐

0.901*** ‐0.734*

[1.159] [1.212] [0.552] [0.669] [0.807] [0.833] [0.327] [0.372]

CPIA score 2.969*** 2.806** 1.139 1.556** 3.349*** 3.449*** 1.625** 1.867***

[1.033] [1.075] [1.082] [0.735] [0.902] [0.939] [0.712] [0.550]

IDA eligible ‐0.402 ‐0.324 4.864*** 5.397*** 1.134 1.162 1.21 1.961***

[2.340] [2.440] [1.194] [0.939] [0.932] [1.006] [0.766] [0.687]

Population (ln) 11.090* 11.003* 1.112*** 0.973*** ‐0.975 0.163 0.998*** 1.027***

[5.727] [5.991] [0.184] [0.353] [2.880] [3.036] [0.138] [0.235]

G3 UNGA‐Alignment ‐ 6.341 ‐ 23.500*** ‐ 7.439* ‐ 10.455***

‐ [7.416] ‐ [4.684] ‐ [4.193] ‐ [2.337]

Total G3 exports (ln) ‐ 0.651 ‐ 0.232 ‐ 0.051 ‐ ‐0.075

‐ [0.492] ‐ [0.311] ‐ [0.372] ‐ [0.216]

G3 colonial heritage ‐ ‐ ‐ ‐0.301 ‐ ‐ ‐ ‐1.027**

‐ ‐ ‐ [0.759] ‐ ‐ ‐ [0.500]

Dependent variable RETF RETF RETF RETF SDTF SDTF SDTF SDTF

Sample All All All All All All All All

Estimation FE FE OLS OLS FE FE OLS OLS

Adjusted R‐squared 0.055 0.057 0.241 0.312 0.56 0.568 0.498 0.534

# Observations 1,375 1,332 1,375 1,332 1,502 1,453 1,502 1,453

# Countries 133 128 133 128 134 128 134 128

Notes: The dependent variables is the natural logarithm of the TF aid noted in the bottom of the respective column

and covers the Bankʹs fiscal years 2002‐2012. All regressions include dummies for the World Bank’s fiscal year. Robust

standard errors in parentheses clustered at the recipient country level. * p<0.10, ** p<0.05, *** p<0.01

27

Table 7: Selectivity with respect to disaster‐ and conflict‐affected countries

1 2 3 4

GNI p.c. (ln) ‐2.448** ‐2.983*** ‐0.528 ‐1.788***

[1.002] [0.854] [0.389] [0.322]

CPIA score 2.210*** 2.213*** 1.255** 1.755***

[0.837] [0.799] [0.555] [0.600]

Population (ln) 3.09 4.262 0.870*** 0.848***

[3.663] [3.042] [0.186] [0.170]

Post‐conflict country ‐0.549 ‐0.238 ‐2.28 ‐0.913

[1.046] [0.940] [4.837] [2.265]

Affected (ln) 0.109 ‐0.011 0.079 0.006

[0.077] [0.067] [0.076] [0.063]

Dependent variable TF TF TF TF

Sample IDA All IDA All

Estimation FE FE OLS OLS

Adjusted R‐squared 0.214 0.157 0.312 0.304

# Observations 768 1,156 768 1,156

# Countries 86 127 86 127

Notes: The dependent variables is the natural logarithm of trust fund aid

and covers the Bankʹs fiscal years 2002‐2012. All regressions include

dummies for the World Bank’s fiscal year. Robust standard errors in

parentheses clustered at the recipient country level. * p<0.10, ** p<0.05, ***

p<0.01

28

Table 8: Sector‐specific selectivity of trust fund aid

1 2 3 4 5 6

GNI p.c. (ln) ‐2.407*** ‐1.614* ‐1.635*** ‐1.328 ‐0.486** ‐0.557

[0.323] [0.973] [0.358] [0.942] [0.245] [0.561]

Population (ln) 0.884*** 14.531*** 1.188*** 10.615** 1.285*** 15.249***

[0.156] [4.508] [0.149] [5.235] [0.137] [2.715]

CPIA, Education 2.047*** 1.647*** ‐ ‐ ‐ ‐

[0.475] [0.547] ‐ ‐ ‐ ‐

CPIA, Health ‐ ‐ 1.017** 0.847 ‐ ‐

‐ ‐ [0.496] [0.614] ‐ ‐

CPIA, Environment ‐ ‐ ‐ ‐ 0.397 ‐0.168

‐ ‐ ‐ ‐ [0.359] [0.357]

Dependent variable TF TF TF TF TF TF

Sample Education Education Health Health Environm. Environm.

Estimation OLS FE OLS FE OLS FE


# Observations 1,367 1,367 1,368 1,368 1,368 1,368

# Countries 132 132 132 132 132 132

Notes: The dependent variable is the natural logarithm of the aid type noted in the bottom of the respective

column and covers the Bankʹs fiscal years 2002‐2012. All regressions include dummies for the World Bank’s

fiscal year. Robust standard errors in parentheses clustered at the recipient country level.

* p<0.10, ** p<0.05, *** p<0.01

poverty and policy selectivity of world bank trust …documents.worldbank.org/curated/en/... ·...

Documents